An ornithologist himself – a neural network will help recognize birds by their singing

Let’s talk about the project, which is used by amateur ornithologists and scientists, and libraries with records of thousands of bird songs – including for commercial use.

Bird spectrograms

There are many research projects devoted to the analysis of bird voices. Thus, scientists seek to assess the population and quality of life of animals in certain regions. It is not uncommon for ornithologists to record long audio recordings of nature sounds, and then are engaged auditory classification.

To automate this process, experts from the Laboratory of Ornithology at Cornell University in the United States developed BirdNET neural network. It identifies over 3,000 bird species based on the spectrogram of their singing. The system places the results on a single cardwhich helps ornithologists keep track of animal populations.

The work of the algorithm can be evaluated on the official project website – it analyzes live audio from a microphone installed in nature. Voices meet in the audio stream red-shouldered troupial, Canadian goose, marsh sparrow and other birds. There is also a demo can be downloaded own audio recording for analysis.

You can work with the neural network not only in the browser. The application runs on both Raspberry Pi and home PC – in particular, under Ubuntu and Windows. Configuration instructions for an open operating system are given in repositories on GitHub. As for Windows, it is enough to download and unpack zip archive and get started as is.

More developers released mobile app. It allows you to record bird singing on a microphone, select the most distinct fragment and send it for recognition. But since the algorithm is susceptible to extraneous noise, it is better to record birds early in the morning or late in the evening.

Who uses the development

The BirdNET neural network has already formed the basis of several research projects. For example, under the initiative BirdWeather enthusiasts are building a map with bird habitats. Dozens of acoustic stations have been installed around the world that monitor natural areas and identify bird voices.

Another project – BirdNET-Pi – allows you to turn a compact Raspberry Pi computer into a mobile ornithological station. The acoustic environment is analyzed using a sound card connected via USB. The authors develop a similar system ecoPi:Bird supported by the German Federal Ministry for the Environment. This is a network of autonomous systems for monitoring bird voices in forests around the country.

More “bird” projects

One of the residents of Hacker News in the topic thread notedthat a large-scale system based on the BirdNET neural network in the future will allow visualizing the migration of birds. However, similar projects already exist – for example, open source system Vesper to monitor the so-called nocturnal flight calls (NFC). These are the sounds that birds make when flying at night to coordinate their movements.

The project is quite young and not ready for large-scale implementation, but in the future it will be able to be used by both amateur ornithologists and biologists.

Another project – Terra. This is a special device designed for installation in the backyard of a country house. It monitors the acoustic background, recognizes the singing of birds living nearby and captures animals with radio tags. Further, the information is entered into a common database for further research.

There are projects that seek not to visualize the movement of birds, but to preserve their voices for posterity. About some of these projects, we told in one of our past articles. Thus, the US National Park Service forms an acoustic library Rocky Mountain Sound Library, in the catalog of which there are already more than two hundred audio recordings of birds and wild animals. By the way, all sounds can be downloaded and freely used for commercial purposes.

A similar set of data is being compiled at the Cornell Lab of Ornithology. In their media library hundreds of thousands of recordings of sounds made by birds, reptiles and insects. All of them are used for research and educational tasks.

What else do we write about in our “World of Hi-Fi”: